Convolutional neural networks for segmenting xylem vessels in stained cross-sectional images

Abstract

Xylem is a vascular tissue that conducts sap (water and dissolved minerals) from the roots to the rest of the plant while providing physical support and resources. Sap is conducted within dead hollow cells (called vessels in flowering plants) arranged to form long pipes. Once formed, vessels do not change their structure and last from years to millennia. Vessels' configuration (size, abundance, and spatial pattern) constitutes a record of the plant-environment relationship, and therefore, a tool for monitoring responses at the plant and ecosystem level. This information can be extracted through quantitative anatomy; however, the effort to identify and measure hundreds of thousands of conductive cells is an inconvenience to the progress needed to have solid assessments of the anatomical-environment relationship. In this paper, we propose an automatic methodology based on convolutional neural networks to segment xylem vessels. It includes a post-processing stage based on the use of redundant information to improve the performance of the outcome and make it useful in different sample configurations. Three different neural networks were tested obtaining similar results (pixel accuracy about 90%), which indicates that the methodology can be effectively used for segmentation of xylem vessels into images with non-homogeneous variations of illumination. The development of accurate automatic tools using CNNs would reduce the entry barriers associated with quantitative xylem anatomy expanding the use of this technique by the scientific community.

Mask, Train, Repeat! Artificial Intelligence for Quantitative Wood Anatomy

“…Our goal was to perform cell instance and lumen area detection together in a single algorithm architecture to maximize the utility for QWA analyses. We treated lumen area detection as an instance segmentation problem rather than a semantic segmentation task as we see in Garcia-Pedrero et al (2019) with the U-Net neural network architecture. The difference between these two is that with the U-Net algorithm every pixel is classified (as “cell” or “not-cell”) independently from each other, whereas in instance segmentation the objects (cells) are first identified and located as a whole, and then segmented to their estimated real dimension in a following stage.…”

“…Wood anatomical research is a field where DCNNs find an ideal application ( Garcia-Pedrero et al, 2019 ). In the past, machine learning methods have mainly been used for wood species identification ( Luis et al, 2009 ; Mallik et al, 2011 ; Ravindran et al, 2018 ; He et al, 2020 ; Wu et al, 2021 ).…”

“…Specifically, we created a practical graphical user interface relying on a Mask-RCNN algorithm architecture ( He et al, 2017 ), that detects cells and quantifies the lumen area on four conifers and three angiosperms. We also present the comparison on the accuracy of cell instance (cell identification as an object) and lumen area detection with another neural network architecture (U-Net; Garcia-Pedrero et al, 2019 ), and ROXAS, one of the most widely used programs for wood anatomical image analyses ( von Arx and Dietz, 2005 ; von Arx and Carrer, 2014 ; Garcia-Pedrero et al, 2019 ).…”

Eigenloss: Combined PCA-Based Loss Function for Polyp Segmentation

“…While U-Net relies on concatenation in the decoder, LinkNet [29] adds the information from the encoder and has also been used for medical image segmentation [30,31] including polyp segmentation [32]. Furthermore, these two networks are usually compared in studies from different medical fields [33][34][35][36]. For polyp segmentation on colonoscopy images, different approaches can be found.…”

Special issue on developing nature-inspired intelligence by neural systems

“…The following paper describes an approach based on convolutional neural networks for the xylem segmentation in stained cross-sectional vessels images [8]. This work shows different quality measure with accuracies between 72 and 92% [8]. There is another paper, which shows a new dataset of sclera biometrics for a public use [9].…”